Sunday, 10 August 2014

Waking the Dead: Exploring the One News Colmar-Brunton Epsom Poll in Comparative Perspective

ACT Party Corpse: Will a nod, a wink and a saucy smile from John Key (but not
 necessarily a cup of tea, this time) be enough to revive an ACT Party on long-term 
life-support ?

National Party Spin-Meister, David Farrar, has one or two rather dodgy things
to say about the latest One News Colmar-Brunton Poll of Epsom voters.

Let's take his musings on the Epsom Party-Vote results, first.

Party Vote

In a Kiwiblog post entitled 'Labour third in Epsom' (Aug 10), Farrar approvingly quotes a One News report on the poll that states: "Six weeks out from the election and the National Party is leading the way in the Epsom electorate......(the) poll...sees National sitting with a comfortable lead in the Epsom electorate, with 60% of those polled in the electorate saying they would vote for National in the election......The Green Party, in second place, trailed far behind with only 16% of Epsom voters saying they would give their party vote to the Greens."

At this point, Farrar very helpfully informs readers that: "Labour were third on 14%", before going on to conclude: "This makes you wonder how many other seats or areas now have Labour in 3rd place for the party vote ? The Greens will be happy, and Labour should be quite worried."

I think we can extract four separate contentions, here, from both Farrar's comments and from the general tone of the One News report:
(1) That National is doing unusually well in Epsom (One News clearly casts National's performance in the Party-Vote section of the poll in highly positive, almost glowing, terms (leading the way, comfortable lead) and Farrar is happy to passively endorse this framing through uncritical quotation)
(2) That the big losers in the poll are Labour (who should be "quite worried") (Farrar regurgitates the Right-leaning meme du jour - that Labour are in danger of being supplanted by the Greens as the major Party of the Left)
(3) That the Greens are doing very poorly (One News emphasise they "trailed far behind" with "only" 16% support)
(4) Conversely, that the Greens are doing quite well (Farrar, in contrast to One News, feels the Greens "will be happy" with their performance in the poll)

The best way to test these contentions is to compare the Party-Vote results of this One News Colmar-Brunton with the actual Party-Vote in Epsom at the 2011 General Election. As you can see from the Table below ((a) Party Vote (Epsom)), contentions (1) and (2) are both wrong. While Labour are certainly down (by 2 points), the biggest losers in the poll relative to 2011, are, in fact, the Nats (down 5 points). National may well be "leading the way" in Epsom and "sitting with a comfortable lead", but - in the context of one of the most Right-leaning seats in the Country - this lead is rather less "comfortable" than it was at the last Election. If Farrar is as eager to extract broader trends and portents from the Epsom poll as he appears to be, then surely he'd be forced to conclude that, in fact, it's National that should be "quite worried" - its nationwide Party-Vote in danger of plunging by 5 points at this year's  General Election (down from 47% to 42%).

More broadly, the Right Bloc are down 4 points in Epsom in this One News Colmar-Brunton, relative to 2011, (down 5 points if you exclude - as the Nats seem to be doing - Colin Craig's Conservatives), while combined Lab+Green support is up 2 points, the Left Bloc (Lab+Green+IMP) is up 3, and the Opposition Bloc (Left Bloc+NZ First) is rating 4 points higher than in the Epsom Party-Vote at the last election. Once again, if we're going to follow Farrar's path and happily extrapolate from this poll then it's certainly not the Left that have cause to be concerned.

As for the Greens individually, Farrar finally gets something right (contention (4)) (though only, of course, because it suits his Greens supplanting Labour meme). The Greens will indeed be pleased with their 4 point rise. One News's contention (3) is thus patently wrong. But unlike Farrar, I'm not at all sure that Labour will be greatly worried by their (relatively close) third place in the poll. As you can see from the 2011 Party-Vote, the percentage point gap between Labour and the Greens in Epsom (as with a number of other affluent Auckland seats) was unusually small. Just 4 points separated the two main parties of the Left in Epsom in 2011, compared to 16 points nationwide. Something Farrar appears to have inadvertently "forgotten" to mention.

I'm a little more wary than Farrar, though, of regarding the party-vote component of this poll as some sort of talisman.  Putting aside the usual caveats about sampling error and the danger of relying on individual polls, it's clear that Epsom is by no means an average or typical seat. Not only is it well to the Right of New Zealand as a whole (National, for example, taking 65% of the Epsom Party-Vote at the last election, compared to 47% nationwide) but also, of course, that unusually close Labour/Green vote. What's more, there are various questions surrounding the voting intentions of the Don't Knows and the likely level of the Non-Vote (24% in Epsom in 2011) which add yet more layers of complexity and uncertainty [Polls generally filter out those respondents who say they probably won't vote - but this unlikely-to-vote component usually comprises a far smaller proportion in opinion polls than Non-Voters do on Election Day].

And yet, then again, I can't help but be struck by just how remarkably similar the party support trends in this One News Colmar-Brunton Epsom poll are to the trends that become obvious when one compares the current batch of nationwide opinion polls with those carried out at the same point before the 2011 Election. I'll be focussing on this in an up-coming post - suffice to say here that in both cases (ie (1) Party Support in Epsom in this latest poll compared to the 2011 Epsom Party-Vote and (2) Party Support in the most recent nationwide opinion polls compared to polling conducted at the same point in 2011)  the Opposition Bloc are up 4 points, the Right Bloc are down 4 (or, excluding the Conservatives, are down 5) and Labour are down 2.  Strong similarities also exist with changes in support for both National (down 4 points in the nationwide polls and down 5 points in the Epsom poll), the Greens (up 3 points and 4 points respectively) and the Left Bloc as a whole (up 2 points and 3 points respectively).(1)

Hence, if Farrar absolutely insists we extrapolate a wider significance from the Party-Vote component of this One News Colmar-Brunton poll of Epsom voters, then the conclusions to be drawn are entirely the inverse of those he's managed (somehow, against the odds) to come up with.

 Table (a):  Party Vote (Epsom)                                                             
 2011 General Election            One News Colmar-Brunton                  

     National      65%                     National      60%            - 5.0            

     Labour        16%                      Labour        14%            - 2.0            
     Green          12%                      Green          16%           + 4.0            
     NZ First      2.6%                     NZ First       3.3%          + 0.7            
     ACT            2.6%                     ACT            2.7%          + 0.1            
     Cons            1.1%                     Cons            2.1%          + 1.0           
     Mana           0.3%                      IMP            1.5%          + 1.2           
     Maori          0.6%                      Maori          0.6%             =              
     Non-Vote     24%                      Don't Know   6%                                

For Candidate Vote analysis, click on Read More

(1) The figures in this paragraph are based on an analysis I completed in mid-August and, hence, are a little out of date now.

Candidate Vote

Epsom, of course, is one of the few seats where the Candidate-Vote has rather more profound electoral consequences than the Party-Vote. Strategic voting by National supporters potentially provides a moribund ACT party, barely registering in the polls, with a crucial life-line. Hence, Candidate-Vote intentions are without doubt the core interest for pollsters in Epsom. Colmar Brunton's modus operandi in this Early August 2014 poll was to ask respondents an initial question: "Thinking about your electorate vote, for your local MP in the Epsom electorate, who would you vote for with your electorate vote ?" (hereafter, I refer to the response to this question as the Initial Response result). Colmar Brunton then informed the entire sample that (i) John Key was encouraging National supporters to give their electorate vote to ACT candidate David Seymour and that (ii) if ACT won Epsom, this might help National form a government after the Election. The pollster then went on to ask the follow-up question: "And with this in mind, who would you now vote for with your electorate vote ?" (hereafter, I refer to this as the Post-Key Prompt or After Key Prompt result). [Note: I also prefer to use the electoral commission's term Candidate Vote rather than Electorate Vote. The former is less ambiguous]

You can see from Table (b) that the Key prompt has quite a profound effect on Candidate-Vote intention. 12 points behind National's Goldsmith in the Initial Response (32% to 44%), ACT's Seymour is the recipient of a pretty hefty swing by National supporters as a result of the Key prompt, ending up a solid 14 points ahead of Goldsmith (45% to 31%). Their positions are essentially reversed.

 Table (b): Candidate Vote (Epsom)   August 2014 Colmar-Brunton   
  (1) Initial Response               (2) After Key Prompt         (3) Diff      
   Goldsmith (Nat)    44%          Goldsmith (Nat)    31%          - 13      
   Seymour  (ACT)    32%          Seymour (ACT)    45%         + 13     
   Wood  (Labour)     10%          Wood (Labour)      9%           - 1       
    Genter (Green)       9%           Genter (Green)     10%          + 1      
    Rankin (Cons)        4%           Rankin (Cons)        4%           =        
    O'Dea (IMP)           1%           O'Dea (IMP)           1%           =       
    Haden (Ind)             0%           Haden (Ind)            0%           =       
    Don't Know             8%           Don't  Know         13%          + 5     

We need to be just a little cautious in interpreting these results, however. Some of the on-line presentations of this Q+A Colmar Brunton have included the Don't Know proportions at the bottom of the results table. And this, in turn, has led a number of commentators on the blogosphere to take the post-prompt support-movement at face-value, erroneously assuming that the candidate percentages are based on the entire sample rather than just the decideds.

Hence, for example, I've seen one or two comments from the Left that (quite understandably) assume that a straight 13 point swing from Goldsmith to Seymour - and a direct 1 point swing from Wood to Genter - occurred as a consequence of the Key prompt. And these commentators have then gone on to criticise Labour supporters for supposedly indulging in a pointless/irrational swing to the Greens' Genter after the prompt. Putting aside the fact that, because we're only seeing the net changes in support, we can't be absolutely sure of where the swings are coming from (although it's a sure bet that the vast majority of the swing to Seymour comes from Goldsmith's initial supporters), the post-prompt movement in support is quite clearly not all it seems.

That's because, despite the inclusion of the Don't Know figures in on-line tables, the Candidate-Vote stats are, in fact, of Decided voters only. And given the significant 5 point increase in the Don't Knows as a result of the Key prompt, a face-value interpretation is inherently misleading.

To see what's really going on, we need to re-calculate on the basis of the entire sample.(2) The results of this re-calculation suggest that, after the Key prompt, Goldsmith's support declines by about 12 points (not too dissimilar to the swing as initially presented), but that support for ACT's Seymour increases by just 9, rather than 13, points. Clearly, while most of that 12 point decline went directly to Seymour, a small minority of Goldsmith's supporters moved into Undecided territory on hearing that Key preferred them to vote strategically.

The re-calculation also suggests that support for Labour's Wood fell by a little more than 1 point after the Key prompt and that support for the Greens' Genter, in fact, rose by less than half a percentage point. It seems quite possible, then, that the major thrust of the swing away from Wood was not to Genter (as some have assumed) but into the Don't Know category. Rather than being criticised for an assumed irrationality, therefore, this minority of Labour supporters probably deserve some credit for presumably deciding to weigh-up a strategic vote in precisely the way their on-line critics would have wished for.

Meanwhile, David Farrar augments his dubious rendition of the Colmar-Brunton Party-Vote results in Epsom with a less-than-kosher interpretation of its Candidate-Vote findings. An interpretation that becomes little more than an advertisement for his readers to Party-Vote ACT. Focussing entirely on the results after the Key-prompt, Farrar assures his readers: "That looks pretty comfortable for David Seymour. ACT have never actually led in a public pre-election poll in Epsom despite winning in 2005, 2008 and 2011. So for them to be ahead six weeks out is a pretty strong signal that they will win the seat." Farrar's ulterior motive then suddenly pops up: "This means that a (Party) vote for ACT will not be a wasted vote. Last time they got 1.1%. If they get 0.1% more and get 1.2% then they get a second MP.
[Incidently, in a recent panel interview for the NZ Herald's 'Election 2014: The Hot Seat' programme, John Key explicitly cited the above argument from Farrar. "Our Pollster", Key suggested, "will tell you he's never seen a poll that's had ACT winning the Epsom electorate and yet for successive years in elections (sic) they've won it." (Farrar's Curia is, of course, the National Party's official pollster)]

There are three reasons why Farrar's analysis won't wash. They are (in order of ascending importance):
(1) No Epsom Polls in 2005 and 2008: Farrar wrongly implies that a whole series of public polls were carried out in Epsom not only at the last election but also in 2005 and 2008. This helps to reinforce the idea that ACT candidate David Seymour's lead in Epsom (in the second (post-Key-prompt) segment of this latest Colmar-Brunton) starkly contrasts with the findings of a large body of Epsom poll data from previous elections. And hence the conclusion that things are looking unusually good for ACT. A fairly extensive on-line search, however, suggests that no public polls whatsoever were carried out in Epsom in 2005 and 2008. Indeed, the only evidence of any poll conducted at either of these two elections comes from an NBR story on an ACT Party Internal poll from 2008 (and, unfortunately for Farrar, that poll suggested ACT's Rodney Hide was well out in front, receiving almost double the support of National MP, Richard Worth. The polar opposite of what we would expect from Farrar's assertions). 
(2) The only 2011 pre-election Epsom poll that (like this latest Colmar-Brunton) prompted respondents about Key's strategic-voting preferences - also, in fact, had the ACT candidate (in this case, John Banks) out in front of National candidate, Paul Goldsmith. So, contrary to Farrar's core argument, the ACT candidate has indeed led in a previous public pre-election poll in Epsom. Just like the current Colmar-Brunton, the 2011 UMR Research poll asked respondents an initial, un-prompted question on who they intended to cast their Candidate Vote for, then went on to inform them that John Key preferred National supporters to vote strategically for the ACT candidate, and followed that up by asking the initial question again in light of this fact. As I'll show in Table (d), in both cases the ACT candidate was, after the Key-prompt, well ahead of National's Goldsmith.
(3) Farrar is by no means comparing Apples with Apples in his analysis. What he does is take the post-Key-prompt figures from the latest Colmar-Brunton and compares them not only with the pre-prompt findings of the 2011 UMR but also with the results of the three other Epsom polls carried out in the 2011 pre-election period, none of which prompted respondents about Key's preference. And, of course, as we've seen, that Key-prompt has a profound effect on voting intention.

So, let's do things properly here in a way that Farrar has conspicuously failed to do. First, in Table (c), I compare the Initial Response findings of the latest Colmar-Brunton with the Initial Response of the 2011 UMR and with the results of the three other 2011 Epsom polls: the Herald on Sunday, the Fairfax, and the Colmar-Brunton of 2011. (I've placed the 2011 polls in chronological order in the Table. The Herald on Sunday (HoS) poll was the first to be carried out in 2011, the Colmar-Brunton (C-B) was the last).

 Table (c) Candidate Vote (Epsom)    (1) Initial Response                        
   2014 Colmar-Brunton   2011HoS   2011UMR  2011Fairfax  2011C-B  
 Nat candidate       44%        57%          42%            46%              41%    
 ACT candidate     32%        33%          28%             29%              30%   
 Lab candidate       10%         8%           19%             15%              17%   
 Green candidate    9%          2%           11%            No Data          11%   
 Undecided             8%          42%          9%               40%         NoData   

As you can see, there's absolutely no basis at all for Farrar's assertion that the ACT candidate is doing unusually well in the latest Epsom Colmar-Brunton poll. Seymour's 32% is only slightly higher than the support Banks received in three of the four 2011 Epsom polls, and slightly lower than Banks received in the other. Similarly, National's Goldsmith in 2014 has very similar Initial Response ratings to his support level in three of the four 2011 polls. His current support is only weak in comparison to one of these four polls - the Herald on Sunday. And even if we focus solely on a comparison with the latter poll, it's clear that the 2014 ACT candidate is by no means the recipient of Goldsmith's 2011 fall in support. (Note, too, the extremely high proportion of Undecideds in the Herald on Sunday and Fairfax polls - the two 2011 polls with the highest support-levels for Goldsmith).  

As I've suggested, the post-Key-prompt figures from the latest Colmar-Brunton should only be compared with the one and only 2011 poll (UMR Research) that, in precisely the same way, prompted respondents by clarifying Key's preferences before re-asking the Candidate Vote question. The comparison is layed-out in Table (d)

Table (d): Candidate Vote (Epsom)    (2) After Key Prompt               
    2014 Colmar-Brunton                                2011 UMR                    
     Goldsmith (Nat) 31%                         Goldsmith (Nat) 25%          
     Seymour (ACT) 45%                              Banks (ACT) 43%          
    Wood (Labour) 9%                              Parker (Labour) 19%         
     Genter (Green) 10%                                  Hay (Green) 11%        
    Undecided 13%                                              Undecided 11%        

Once again, you can see that Farrar's contention holds no water. After the Key-prompt, ACT's Seymour attains an only slightly higher level of support than Banks did following the same prompt in 2011 (45% compared to 43% respectively). What's more, National's Goldsmith is also up in 2014 and by a larger proportion - 31% (2014) compared to 25% (2011). In other words, the ACT candidate of 2011 (Banks) led National's Goldsmith by more (18 points) than ACT's Seymour does now (14 points). Hard to see how this puts Seymour in a position of unprecedented strength.

There's also the question of just how accurate pre-Election Epsom polls have been. As Table (e) shows, Banks in fact managed to squeak in by a much narrower margin on Election Day 2011 than the (post-Key prompt) findings of the 2011 UMR poll would have had us believe. True, Banks' support rose slightly (by one point - from 43 to 44%), but Goldsmith's rose much more (up 13 points - 25 to 38%). Clearly, a swathe of Labour and Green voters had decided in the final few weeks of the campaign to hold their collective noses and vote strategically for Goldsmith in order to throw ACT out of Parliament and hence make the chances of a National-led Government rather more precarious.

But, "at the end of the day", although Farrar is (i) entirely misleading on the significance of the party-vote results and although he (ii) employs a decidedly dodgy comparative approach in order to gild the lily when it comes to Seymour's pre-Election chances, I would have to agree that the ACT candidate is still odds-on favourite to hold the seat. Despite concerns expressed by one or two Left-leaning commentators to the contrary, it looks to me like Labour and Green Party-voters in Epsom are indicating at a much earlier stage than in 2011 that they're prepared to vote strategically for the National candidate. 

In the 2011 UMR, 19% of respondents said they'd party-vote Labour and precisely the same proportion said they'd candidate-vote Labour. For the Greens, the  figures are 10% and 11% respectively. So at the time of the 2011 pre-election UMR, it seems that the vast majority of Labour and Green supporters were intending to candidate-vote Left (taking into account, of course, the fact that some Parker and Hay voters would have been supporters of other parties).

In the event, 35% of Labour Party-voters and 54% of Green Party-voters ended up strategically casting their candidate-vote for Goldsmith on Election Day 2011. What's important here, though, is the fact that most of them weren't thinking of voting strategically a few weeks out, at the time of the UMR. In stark contrast, the 2014 Q+A Colmar Brunton suggests that, by early August this year, a fairly healthy minority of Labour and Green voters were already signalling a desire to vote strategically (14% of the entire sample favoured Labour - and 16% the Greens - with their Party-Vote, but only 9% and 10% of all respondents were thinking of Candidate-voting for those two parties respectively). That, in turn, suggests the 14 point gap between Seymour and Goldsmith probably won't narrow to the extent that it did in 2011 (ie Goldsmith's Candidate-Vote result in the 2014 Colmar Brunton already partially includes the strategic swing in sentiment by Labour and Green supporters that, in 2011, had only occurred after the UMR was conducted).

All of which means that Farrar could have come to a not entirely dissimilar conclusion without having to resort to a strategy involving intellectual sleight-of-hand.

Comprising a sizeable 60% of the Epsom electorate, National Party-voters are really the key here. But, much like chocolate or bloodstains on a business shirt in the wash, they're going to be hell to shift. As I said back in June (The Ides of Epsom post), "We need to remember that even in the wake of various scandals that enveloped ACT before the 2011 Election, 60% of Epsom Nats were still prepared to hold their nose and cast their candidate-vote for Banks."

  Table (e):  Candidate Vote  (Epsom)                                               
              2011 UMR                       2011 General Election                
     Goldsmith (Nat)  25%                Goldsmith (Nat)  38%               
      Banks (ACT)     43%                      Banks (ACT) 44%               
        Parker (Lab)    19%                     Parker (Lab)   11%                         
       Hay (Green)      11%                       Hay (Green)   6%               
        Undecided        11%